Can AI Translate Gel Formulas into Predictable Strength? Data‑Driven Design of Super‑Adhesive Hydrogels

Background

Hydrogels are widely used in medical devices, tissue engineering and biomimetic materials, but small changes in formulation cause large, unpredictable variations in mechanical properties such as strength, elasticity and toughness.

Data‑driven formulation workflow

The authors assembled a database of 24,707 adhesive proteins from archaea, bacteria, eukaryotes, viruses and synthetic sources, covering 3,822 species. After ranking species by protein count, the top 200 were selected for detailed analysis.

From these proteins six functional sequence motifs were identified. Each motif was mapped to a synthetic monomer that mimics the corresponding amino‑acid functionality. The six monomers were used as building blocks in a Monte Carlo simulation based on the Mayo–Lewis copolymerization model, generating 180 distinct copolymer sequences.

Experimental validation

All 180 gels were synthesized and mechanically tested for underwater adhesive strength (Fₐ). Sixteen formulations exceeded 100 kPa, 83 exceeded 46 kPa, and the best performer (designated G‑max) reached 147 kPa.

Machine‑learning‑guided optimization

A batch sequential model‑based optimization (SMBO) loop was applied to the experimental data to propose new formulations. Three SMBO‑derived candidates—R1‑max, R2‑max and R3‑max—showed network topologies similar to G‑max but with improved properties after iterative ML refinement.

Key compositional insights:

Inclusion of the hydrophobic monomer BA and the aromatic monomer PEA reduces interfacial water and enhances physical contact with substrates.

The cationic monomer ATAC provides electrostatic adhesion to negatively charged surfaces (e.g., glass). Excess ATAC leads to over‑swelling and reduced contact, so its fraction must be balanced.

Performance of the optimized gels

R1‑max achieved adhesive strength >1 MPa on glass, retained stable adhesion over >200 attach‑detach cycles, and withstood a 1‑kg shear load for more than one year. In seawater, R1‑max held a rubber duck on a rock against tides. R2‑max sealed a 3‑m‑high, 20 mm‑diameter polycarbonate pipe filled with tap water. R3‑max displayed comparable performance.

Limitations and outlook

The current monomer library is limited and precise control of polymer sequence during synthesis remains challenging. Expanding the modular monomer set, improving polymerization techniques, and developing more generalizable physics‑informed ML models will broaden the applicability of the workflow to conductive, responsive or degradable hydrogels for underwater robotics, wound‑care patches and soft‑organ repair.

Paper link: https://www.nature.com/articles/s41586-025-09269-4

Code example

来源：ScienceAI
本文
约2000字
，建议阅读
5
分钟
能不能用 AI 做个「配方翻译器」：给它一组胶配方，它告诉你强度如何、延展性几何；甚至反过来，根据你想要的性能，AI 推荐配方？